My agent has a face
In less than two months, three new agents have arrived, and two of them have a face. Meta launched Muse in early September, and at its Connect event it gave it an avatar: a cream-colored doll with black eyes and a smile, called Jolly. OpenAI answered at its DevDay with Dots, small colorful creatures you can give a name to. And in August, xAI presented its Grok Bots, agents that work like teammates inside a chat.
When I saw Jolly, I thought of two things: the Tamagotchi of the 90s and today’s Labubu craze. I’m not the only one: several media outlets have already compared it to a Tamagotchi, and others to Clippy, the famous Office paperclip.
In this article I want to explain why I think this is not a coincidence. The more “machine” technology becomes, the more we need it to look like something human. And that changes what we designers will design in the coming years.
When we force people to talk like machines, the opposite happens
The first personal computers forced us to speak their language. You had to type commands, remember instructions and accept errors nobody understood. For most people, it was not a natural relationship.
The answer was not to teach everyone to code. It was to make the machine look like our world. In the 80s, Susan Kare designed the icons of the first Macintosh: the trash can, the folder, the document, the clock, the smiling Mac. Objects we already knew from the office. Suddenly, you could understand a computer without a manual. Years later, Steve Jobs took this idea further with skeuomorphism: the wooden bookshelf of iBooks on the iPad and iPhone looked like a real bookcase.
This is the idea I want to defend: every time technology takes us away from what is human, design ends up finding the way back. First it was icons. Then touch screens. Now it’s voice, conversation and, once again… a face?
Screens are disappearing
As Javier Cañada (opens in a new tab), director of Instituto Tramontana, says in the Spanish podcast El fin de la interfaz (opens in a new tab) (“The end of the interface”), it looks like UI screens are going to disappear and everything will become conversational. The new agents already point in that direction. They don’t have a dashboard full of buttons: you write or talk to them, and they do the work.
Also, they no longer live only in their own app. They live where we already talk to other people:
- Muse can be used inside WhatsApp chats.
- Dots can be used from Slack and Microsoft Teams, and OpenAI is also preparing text messages.
- Grok Bot works like a messenger. You can put several bots in the same chat, where they talk to each other and split the tasks.
The agent is no longer a tool you open. It’s someone else in the conversation, like you and me in a Telegram group or a Slack channel. And if it’s “someone”, it makes sense that it ends up having a name, a personality and a “mug”.
The risk of a machine that is too human
This is what worries me most. Today, some people use chatbots as friends or even as therapists. Synthetic voices are already very convincing: they pause, change their tone and sound empathetic. It’s easy to forget that there is nobody on the other side.
For a designer, this is a responsibility. Making something more human makes it easier to use, but also easier to believe.
Human, but only up to a point
Robots look like humans, but only up to a point. The uncanny valley (opens in a new tab) explains it well. It’s an idea from the Japanese roboticist Masahiro Mori: the more a robot looks like a person, the more we like it. But when it gets very close to human without quite getting there, our liking drops suddenly and we feel rejection. It starts to get creepy!
That’s why Figure AI’s robots have a human body but don’t try to have a human face. They walk and use their hands like us, but nobody confuses them with a person. It’s a design decision, not a technical limitation. The same goes for robot dogs, like Boston Dynamics’ Spot, which doesn’t even have a real head.
Jolly and the Dots follow the same logic. They are not realistic faces. They are characters: simple eyes, round shapes, soft colors, and fuzzy “little hairs” to make them more huggable. They stay on the safe side of the valley, like a plush toy or a Tamagotchi.
An agent friend with a face and eyes
After the bots that join our chats, Muse arrives to tell us that our agent friend has a face and eyes we can talk to. Meta announced that every user will be able to create their own avatar for their agent and talk to it in real-time video calls. For me, the most interesting part was the Muse Charm, a small device to talk to your agent as if it were a Tamagotchi 2.0, hanging from your keys. We already saw the Rabbit R1 (and its failure), but now Meta has given the idea a new twist.
Right or wrong, I think the idea is not all bad. Humans have always liked having objects to touch and look at. A Tamagotchi was just a small screen with a few pixels, but people looked after it like a pet. Labubu dolls don’t do anything, and people still queue to buy them and hang them from their bags. We need something physical to care about.
The duck and the decorated shed
In architecture school, there is a book almost all of us end up reading: Learning from Las Vegas (1972), by Robert Venturi, Denise Scott Brown and Steven Izenour. In it, they describe two types of buildings.
The first one is the duck. The name comes from a real shop on Long Island that sold ducks and eggs, and that was shaped like a giant duck. The building is its message: you don’t need any sign to know what’s inside.
The second one is the decorated shed: a normal, functional and cheap building, with a big sign in front that explains what it is. The shape says nothing; the message is on the sign.
For years, software has been a decorated shed. Inside, almost all apps look alike: lists, buttons, menus. The brand was the logo and the colors, the sign hanging on the facade.
With agents, something new happens. If there is no screen, there is no facade to hang the sign on. That’s why Meta and OpenAI are building ducks: Jolly and the Dots are the product. Their shape, their voice and their character have to explain, without a manual, what the agent is and what you can expect from it.
Venturi, by the way, defended the decorated shed. But with agents, that option is starting to disappear. And here is the responsibility: a duck promises what’s inside. If the agent looks like a sweet, friendly plush toy, people will expect it to behave like one. The more the shape promises, the bigger the disappointment when the product doesn’t live up to it.
Interfaces disappear, objects appear
Screens may disappear, but design doesn’t. It just moves somewhere else. If the agent no longer needs buttons or menus, we will design other things: its voice, its character, its face and the physical objects that carry it.
If the agent is a duck, its shape is also design. And just like in architecture, we don’t only design a function: we design something people will live with every day, in their home, in their pocket or hanging from their keys.
Susan Kare used a trash can and a folder to help us understand the computer. Today, Meta, and even OpenAI, use something like a plush toy. The question for designers is not whether our agents will have a face. It’s how to design that face so it helps us understand them, without making us believe they are something they are not.